CN109800772A - A kind of data identification method - Google Patents
A kind of data identification method Download PDFInfo
- Publication number
- CN109800772A CN109800772A CN201910092929.4A CN201910092929A CN109800772A CN 109800772 A CN109800772 A CN 109800772A CN 201910092929 A CN201910092929 A CN 201910092929A CN 109800772 A CN109800772 A CN 109800772A
- Authority
- CN
- China
- Prior art keywords
- image
- value
- color
- classification
- disaggregated model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Landscapes
- Image Analysis (AREA)
Abstract
The present invention relates to field of image recognition, and in particular to a kind of data identification method, comprising: image acquisition step acquires the image of object;Characteristic value calculates step, and the color characteristic of image is extracted from the image of acquisition, calculates the color feature value in color of image feature;The color feature value of image is input to by exporting the classification results of image in disaggregated model trained in advance by image classification step;Image classification step further include: classification thresholds Optimization Steps, it obtains sample image and calculates color feature value, the color feature value of input sample image obtains initial value to disaggregated model, the actual value of initial value and sample image is carried out error amount is calculated, inversely the classification thresholds of disaggregated model are optimized using error amount, until error amount is less than preset minimum value.The present invention can solve existing disaggregated model because using the problem that versatility caused by fixed cluster threshold value is low and classification is inaccurate.
Description
Technical field
The present invention relates to field of image recognition, and in particular to a kind of data identification method.
Background technique
Ceramic tile is very widely used building decoration product, and it is more that a large amount of such as calcium, magnesium, potassium, sodium are contained in ceramic tile formula
Kind mineral raw material, is exactly these mineral raw materials (namely reasonable firing temperature) in specific sintering process, can
Melting merge become not only it is hard but also it is resistance to freeze, be wear-resisting, corrosion resistant Imitation Rock Porcelain Tiles.But these minerals melt merge when, no
Same temperature can show different colors, and this generates color difference.In the prior art using manually to the porcelain for having color distinction
Brick is classified, and the control of the effect and speed of classification by the mood of people causes to generate many unstable factors.
To solve the above problems, the Chinese patent of Publication No. CN203061453U discloses a kind of ceramic tile color difference vision point
Select system, including transmission belt, optical flip-flop and detection case;The optical flip-flop and the detection case are installed in described
On the coil holder of transmission belt;Light source, industrial camera, image pick-up card and industrial personal computer are installed in the detection case;The industry control
Vision-based detection and check and evaluation software systems are configured in machine;The industrial camera respectively with the optical flip-flop, the figure
As capture card connection, and the optical flip-flop, described image capture card are connect with the industrial personal computer.The program can be to presence
The ceramic tile of color difference makes automatic sorting.
During sintering, color can change ceramic tile with variation with temperature, and strictly speaking each temperature can
A kind of color is corresponded to, it is only minimum adjacent to color difference, visually it is difficult to distinguish.Ceramic tile is in sintering, it is difficult to ensure that ceramic tile
Each position is heated consistent, and therefore, the ceramic tile of different batches color under identical sintering condition also has a deviation, and existing skill
When disaggregated model in art is identified and classified to ceramic tile, fixed classification thresholds are all made of, when there is color error ratio
It cannot adjust, cause versatility low and the problem of classification inaccuracy.
Summary of the invention
The purpose of the present invention is to provide a kind of data identification methods, are avoided that existing disaggregated model because using fixed cluster
The problem that versatility caused by threshold value is low and classification is inaccurate.
A kind of base case provided by the invention are as follows: data identification method, comprising the following steps:
Image acquisition step acquires the image of object;
Characteristic value calculates step, and the color characteristic of image is extracted from the image of acquisition, calculates the face of color of image feature
Color characteristic value;
The color feature value of image is input to by being obtained in disaggregated model trained in advance first by image classification step
To current value, then current value is compared with preset classification thresholds, the classification results of final output image;
Classification thresholds Optimization Steps obtain sample image and calculate the color feature value of sample image, input sample image
Color feature value obtain initial value to disaggregated model, the actual value of initial value and sample image is carried out error is calculated
Value, inversely optimizes the classification thresholds of disaggregated model using error amount, until error amount is less than preset minimum value.
Beneficial effects of the present invention: this programme classifies to image by disaggregated model, while can calculate error amount, and
Real-time optimization is carried out by classification thresholds of the error amount to disaggregated model, can solve existing disaggregated model using fixed cluster threshold value
The problem that caused versatility is low and classification is inaccurate.
It further, further include train classification models step, first using color feature value as training characteristics value, according to training
The feature of image is normalized in characteristic value, then determines the structure and network initial weight of disaggregated model, is calculated using heredity
The initial weight of method Optimum Classification model is finally obtained and has been trained using normalized characteristic train classification models network
At disaggregated model.
This programme is classified using disaggregated model, and the initial weight of disaggregated model uses genetic algorithm optimization, the present invention
By using in training set characteristics of image and classification results prototype network is trained, obtained feature and classification results
Mapping relations.
The utility model has the advantages that classifying in conjunction with new feature and blending algorithm, operation is iterated by gradient algorithm and solves power
The process of value continuously adjusts the weight and threshold value of network by training, so that the error of the current value of output and actual value
It is small, so that the input-output mappings relationship for keeping network implementations given, significantly improves nicety of grading.
Further, characteristic value calculates in step, and the color feature value for calculating image specifically includes: pre-treatment step, to figure
As carrying out edge detection, edge pixel point is obtained, first area is determined according to edge pixel point;Step is calculated, image RGB is empty
Between all colours value be converted to the color value in the space LAB, then calculate the Euclideans of any two kinds of color values in first area away from
From.
The utility model has the advantages that color imperceptible for human eye, most of marginal portion for being all present in image, it is therefore desirable to
It detects the edge pixel point of image to be the advantages of carrying out subsequent processing, RGB is converted to LAB: computer knowledge can be reduced
The difficulty of other color allows algorithm to identify color in the less space of color.
Further, characteristic value calculates in step, further includes noise reduction step before pre-treatment step, carries out at noise reduction to image
Reason.
The utility model has the advantages that can be reduced noise in digital picture by noise reduction can prevent the interference of external environmental noise.
Further, characteristic value calculates in step, and the color of image feature of extraction includes number of greyscale levels, color similarity, face
Color gradual change and saturation degree.
The invention also discloses a kind of data recognition systems, comprising:
Image capture module, for acquiring the image of object;
Image processing module, the color for obtaining the target object image of image capture module acquisition and extract image are special
Sign calculates the color feature value of image according to color characteristic;
Image classification module, the color feature value calculated for image processing module and by color feature value and classification thresholds
Compare, judge the classification of image according to the result of the comparison and obtains the classification results of image.
Further, further includes: disaggregated model training module, for using the tile image of known classification results as training sample
This train classification models.
The utility model has the advantages that this programme forms disaggregated model in the way of to tile image progress machine learning, and then sharp again
It uses the tile image of shooting to classify as object of classification, using the mode classification of artificial intelligence, eliminates traditional artificial
Method for separating due to variation, worker by light physiology and psychology variation, sight angle variation etc. factors influenced, and
And homogeneous classification standard improves the accuracy for causing the color difference of entire process to sort.
Detailed description of the invention
Fig. 1 is the flow diagram of data identification method in the embodiment of the present invention one;
Fig. 2 is the logic diagram of ceramic tile cutting system in the embodiment of the present invention two.
Specific embodiment
Below by the further details of explanation of specific embodiment:
Embodiment one:
A kind of data identification method includes: as shown in Figure 1:
S1: image acquisition step acquires the image of object.
Adopting for image is completed using the image of constant enlarged DH-HV300 type colour technical grade video camera shooting ceramic tile surface
Collection, this video camera have been internally integrated A/D conversion module and can directly have been handled by USB2.0 interface with PC machine.
S2: characteristic value calculates step, and the color characteristic of image is extracted from the image of acquisition, calculates in color of image feature
Color feature value.
The extraction of color of image feature is carried out using direct figure method: on the basis of determining space, statistics each color point
The pixel of amount accounts for the ratio of the total pixel of image, obtains the ratio distribution (histogram) of image various colors, finally histogram is made
Image retrieval is carried out for the color characteristic of image.
S201: noise reduction step carries out noise reduction process to image.
S202: pre-treatment step carries out edge detection to image, obtains edge pixel point, determined according to edge pixel point
First area.Realize that edge detection, color imperceptible for human eye are most of using existing technique of image edge detection
All it is present in the marginal portion of image, it is therefore desirable to detect the edge pixel point of image to carry out subsequent processing.
S203: calculating step, all colours value of image rgb space be converted to the color value in the space LAB, then the
The Euclidean distance of any two kinds of color values in one region.
After extraction obtains color of image feature, the color feature value of image is calculated by existing PC software: by image RGB
The color value of all colours in space is converted to the color value in the space LAB, and the pixel number of each color is constant after conversion.LAB
Color space is than rgb color space closer to human vision.It is calculated, can be expired on rgb color space to greatest extent
The premise of sufficient color identification, and can recognize that the number of colours for meeting eye recognition.
In non-edge pixels point region, any two kinds of face in the space LAB is calculated according to the sequence of color value from more to less
The distance of the color value of color.Any two kinds of colors in first area are calculated according to the sequence of color value from more to less according to formula
The distance of LAB value (L1*, a1*, b1*) and (L2*, a2*, b2*), wherein L1*, a1*, b1* and L1*, a1*, b1* are respectively two
The value in three channels of the LAB color space of kind color.
S3: the color feature value of image is input in the disaggregated model by training in advance, this reality by image classification step
Applying disaggregated model in example is BP neural network model, obtains current value first, is then compared current value with classification thresholds,
The classification results of final output image.
S301: training BP neural network model step, input sample color feature value instruct BP neural network model
Practice.
Training BP neural network model: first using color feature value as training characteristics value, according to training characteristics value to figure
The feature of picture is normalized, and then determines the structure and network initial weight of BP neural network model, excellent using genetic algorithm
Change BP neural network initial weight, using normalized characteristic training network, obtains trained BP neural network model.
S302: the color feature value of target object image is input in BP neural network model and obtains by image classification step
Current value is compared by the current value of image with classification thresholds, completes the classification to target image.
In the present embodiment, the color difference situation of image is divided into no color differnece, slight color differences, obvious color difference and serious color difference,
The classification thresholds range of middle no color differnece is 0-0.5, and the classification thresholds range of slight color differences is 0.5-1.0, hence it is evident that the classification of color difference
Threshold range is 1.0-1.5, and the classification thresholds range of serious color difference is 1.5-2.0.
It is compared according to the current value of target image with classification thresholds, for example, the current value of certain target image is 0.72,
So the target image is divided into slight color differences one kind.
S303: classification thresholds Optimization Steps obtain sample image and calculate color feature value, the color of input sample image
Characteristic value obtains initial value to BP neural network model, the actual value of initial value and sample image is carried out error is calculated
Value, inversely optimizes the classification thresholds of BP neural network model using error amount, until error amount is less than minimum value.
Optimized BP Neural Network model: sample of color characteristic value is supplied to input neuron, in BP neural network model
Portion successively propagates forward signal, finally generates the initial value of output layer, carries out initial value and actual value that error is calculated
Error amount is inversely traveled to hidden neuron if error amount is greater than minimum value by value, according to error amount come adjusting and optimizing nerve
The connection weight of member and biasing and classification thresholds, then the color characteristic of input picture continues to train, until initial value and reality
Error amount between actual value is less than minimum value.In the present embodiment, minimum value 0.02.
Such as: the color feature value of first sample is R=25, G=56, B=48, and color feature value is input to BP nerve
The initial value exported in network model is 0.52, but the actual value of first sample is 0.75, then error amount is exactly 0.23, it will accidentally
Difference inversely travels in hidden neuron, being adjusted and optimize to BP neural network model, until inputting first sample
Color characteristic to rear output initial value be 0.5.
The color feature value of second sample is R=82, G=109, B=115, which is input to BP nerve
The initial value exported in network model is 0.93, but the actual value of the second sample is 1.00, then error amount is 0.07, by error
Value inversely travels in hidden neuron, being adjusted and optimize to BP neural network model, until inputting the second sample
The initial value of color feature value output is 1.00.
The color feature value of third sample is R=154, G=188, B=210, which is input to BP nerve
The initial value exported in network model is 1.81, and the actual value of third sample is 1.80, and error amount 0.01 is less than minimum,
0.01, complete the training to BP neural network model.
The present embodiment also discloses a kind of data recognition system, including server and acquisition terminal, and server and acquisition are eventually
End is connected by existing wireless connection module network.
Acquisition terminal, comprising:
Image capture module, for acquiring the image data of object.
The acquisition of image is realized using existing video camera and optical lens.In this implementation, using constant enlarged DH-HV300
Type colour technical grade video camera shoots the image of ceramic tile surface to complete the acquisition of image, this video camera has been internally integrated A/D
Conversion module can be handled directly by USB2.0 interface with PC machine.The resolution of video camera of this model is high, highest resolution
Up to 2048x1536,5V is used to power, line is simple.Optical lens is using Japanese computer camera lens.
Server, comprising:
Image processing module, for obtaining the object image data of image capture module acquisition and extracting image data
Color characteristic calculates the color feature value of image according to color characteristic.
Image classification module, the color feature value calculated for image processing module and by color feature value and classification thresholds
Compare, judge the classification of image according to the result of the comparison and obtains the classification results of image.
Disaggregated model training module, for the training classification mould using the tile image of known classification results as training sample
Type.
Server be existing PC machine, wherein the function of image processing module and image analysis module by PC machine hardware and
Software is realized in conjunction with existing program.Wherein the configuration of PC machine is as follows: CPU: four processor of Pentium;Mainboard: prestige star MSI865PE;
Memory: Jin Shidun Kingston512DDE400;Hard disk: Seagate ST80G/7200;Video card: seven rainbow FX5200/128M;CD-ROM drive:
LG16CxDVD.All image datas are all directly to call in memory processing, processing analysis relatively high to the request memory of computer
Picture format be all BMP format, the file of BMP format is inherently bigger, memory use 1G memory (two Jin Shidun
512M)。
Embodiment two:
The present embodiment the difference is that only that the server in the present embodiment further includes that ceramic tile is cut out compared with embodiment one
Cut system.
The color difference of ceramic tile is divided into whole color difference and local color difference, the ceramic tile of whole color difference occurs, except carry out color-match (with
The ceramic tile of similar color is matched) outside, waste disposal can only be regarded;For the ceramic tile of local color difference, color difference part can be cut
It removes, ceramic tile is cut to the smaller ceramic tile of overall dimensions.How to realize optimal cutting (part that excision abandons is minimum) or incites somebody to action
It is urgent problem that color difference part, which rationally utilize,.
It is as shown in Figure 2: ceramic tile cutting system, comprising:
Pattern colour difference division module, the image for having color difference sub-elected for receiving image classification module, and according to image
Color feature value calculate image color difference point, identified by color difference point coordinatograph and to the boundary coordinate of color difference point, root
Subregion is carried out to image according to the boundary of color difference point and obtains color difference block and no color differnece block.
It is identified using color feature value of the existing image-recognizing method to image, wherein pattern colour not good enough boundary
" a kind of edge detection of the color image based on color difference disclosed in School of Information Technology, East China University of Science can be used in identification method
Method " in the edge detection method of image identified, complete that the part for generating color difference is defined as color difference after Boundary Recognition
The part of no color differnece is defined as no color differnece block by block.
Image cropping module, for obtaining the image after pattern colour difference division module subregion, firstly, according to the side of color difference point
Boundary calculates the shape and area of non-color difference block, obtains the size of small tiles, according to the size of multiple identical or different small tiles
Splicing cuts non-color difference block to obtain small porcelain block and rest block, then, calculates the shape and area and will be remaining of rest block
Block is cut into the maximum rectangle patch of area, finally, obtaining mould according to the shape and area of the feature modeling color difference block of color difference point
The shape and area of Cutting model in type database and the cutting mould that maximum area is selected according to the shape of color difference block and area
Type is cut, and ornaments block is obtained.
By the boundary that existing image recognition technology scans color difference block and non-color difference block can be calculated color difference block and
The shape of non-color difference block can calculate the area of color difference block and non-color difference block by AutoCAD: input order AREA, selected object,
Obtain area.Object block cutting can be calculated by way of exposure mask, BuildMask tool is used in the present embodiment: being obtained
The dimension information of small tiles constructs masked areas, is then multiplied masked areas with non-color difference block image to obtain small porcelain block.
It is previously stored with art work Cutting model, including plant: the moulds such as fresh flower, leaf, bunge bedstraw herb, bouquet in the database
Type;Animal: the models such as doggie, piggy, dinosaur;Cartoon character: the models such as Lu Fei, ring people, Hanamichi Sakuragi.
Small porcelain block is the ceramic tile that color is identical as object but specification (size) is different, and patch and ornaments block can be used as present
It is distributed to user, wherein small porcelain of the patch for cover when rest position is not enough to install whole tile when installing ceramic tile
Brick, ornaments block can be used as ornaments.
The present embodiment the utility model has the advantages that 1) by calculate the non-color difference block of ceramic tile shape and size, can be to there are color difference
Ceramic tile is cut, and smaller size of ceramic tile and the patch for cover are obtained, can be real maximumlly using the ceramic tile for having color difference
The recycling for having showed substandard products ceramic tile solves the problems, such as waste material and productivity caused by directly scrapping processing;2) for ceramic tile
On there is color difference part maximize and cut, used as ornaments, the present embodiment realizes optimal cutting and will be coloured
The effect that poor part is rationally utilized.
What has been described above is only an embodiment of the present invention, and the common sense such as well known specific structure and characteristic are not made herein in scheme
Excessive description, technical field that the present invention belongs to is all before one skilled in the art know the applying date or priority date
Ordinary technical knowledge can know the prior art all in the field, and have using routine experiment hand before the date
The ability of section, one skilled in the art can improve and be implemented in conjunction with self-ability under the enlightenment that the application provides
This programme, some typical known features or known method should not become one skilled in the art and implement the application
Obstacle.It should be pointed out that for those skilled in the art, without departing from the structure of the invention, can also make
Several modifications and improvements out, these also should be considered as protection scope of the present invention, these all will not influence the effect that the present invention is implemented
Fruit and patent practicability.The scope of protection required by this application should be based on the content of the claims, the tool in specification
The records such as body embodiment can be used for explaining the content of claim.
Claims (7)
1. a kind of data identification method, it is characterised in that: the following steps are included:
Image acquisition step acquires the image of object;
Characteristic value calculates step, and the color characteristic of image is extracted from the image of acquisition, and the color for calculating color of image feature is special
Value indicative;
The color feature value of image is input to by being worked as in disaggregated model trained in advance first by image classification step
Then current value is compared, the classification results of final output image by preceding value with preset classification thresholds;
Classification thresholds Optimization Steps obtain the sample image of known classification results and calculate the color feature value of sample image, defeated
The color feature value for entering sample image obtains initial value to disaggregated model, and the actual value of initial value and sample image is calculated
Obtain error amount, inversely the classification thresholds of disaggregated model optimized using error amount, until error amount be less than it is preset most
Small value.
2. data identification method according to claim 1, it is characterised in that: further include train classification models step, first
It using color feature value as training characteristics value, is normalized according to feature of the training characteristics value to image, then determines classification
The structure and network initial weight of model, using the initial weight of genetic algorithm optimization disaggregated model, using normalized feature
Data train classification models network finally obtains the disaggregated model of training completion.
3. data identification method according to claim 2, it is characterised in that: characteristic value calculates in step, calculates image
Color feature value specifically includes: pre-treatment step, carries out edge detection to image, edge pixel point is obtained, according to edge pixel
Point determines first area;Step is calculated, all colours value of image rgb space is converted to the color value in the space LAB, is then counted
Calculate the Euclidean distance of any two kinds of color values in first area.
4. data identification method according to claim 3, it is characterised in that: characteristic value calculates in step, pre-treatment step
Before further include noise reduction step, to image carry out noise reduction process.
5. data identification method according to claim 4, it is characterised in that: characteristic value calculates in step, the image of extraction
Color characteristic includes number of greyscale levels, color similarity, color gradient and saturation degree.
6. a kind of data recognition system, it is characterised in that: include:
Image capture module, for acquiring the image of object;
Image processing module, for obtaining the target object image and the color characteristic for extracting image that image capture module acquires, root
The color feature value of image is calculated according to color characteristic;
Image classification module, the color feature value calculated for image processing module and by color feature value and preset classification threshold
Value compares, and judges the classification of image according to the result of the comparison and obtains the classification results of image.
7. data recognition system according to claim 6, it is characterised in that: further include:
Disaggregated model training module, for using the tile image of known classification results as training sample train classification models.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910092929.4A CN109800772A (en) | 2019-01-30 | 2019-01-30 | A kind of data identification method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910092929.4A CN109800772A (en) | 2019-01-30 | 2019-01-30 | A kind of data identification method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109800772A true CN109800772A (en) | 2019-05-24 |
Family
ID=66560578
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910092929.4A Pending CN109800772A (en) | 2019-01-30 | 2019-01-30 | A kind of data identification method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109800772A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111134613A (en) * | 2019-11-21 | 2020-05-12 | 明灏科技(北京)有限公司 | Image recognition-based orthokeratology lens fitting method and system |
CN112487950A (en) * | 2020-11-27 | 2021-03-12 | 上海建工四建集团有限公司 | Intelligent recognition system and method for defects of bare bricks |
CN113135368A (en) * | 2021-04-29 | 2021-07-20 | 怀化学院 | Intelligent garbage front-end classification system and method |
CN115434054A (en) * | 2022-10-28 | 2022-12-06 | 富尔美技术纺织(苏州)有限公司 | Multicolor fiber uniform mixing processing technology |
CN115937085A (en) * | 2022-06-28 | 2023-04-07 | 哈尔滨学院 | Nuclear cataract image processing method based on neural network learning |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102663451A (en) * | 2012-03-29 | 2012-09-12 | 天津科技大学 | Graph image classification method based on color space characteristics |
CN203061453U (en) * | 2012-10-31 | 2013-07-17 | 张伟群 | Vision separation system for color difference of tiles |
CN103593511A (en) * | 2013-10-27 | 2014-02-19 | 沈阳建筑大学 | Cutting-off and layout method for natural stone slabs with defects |
CN105430218A (en) * | 2015-11-27 | 2016-03-23 | 优渊商贸(上海)有限公司 | Image color recognition method and device |
-
2019
- 2019-01-30 CN CN201910092929.4A patent/CN109800772A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102663451A (en) * | 2012-03-29 | 2012-09-12 | 天津科技大学 | Graph image classification method based on color space characteristics |
CN203061453U (en) * | 2012-10-31 | 2013-07-17 | 张伟群 | Vision separation system for color difference of tiles |
CN103593511A (en) * | 2013-10-27 | 2014-02-19 | 沈阳建筑大学 | Cutting-off and layout method for natural stone slabs with defects |
CN105430218A (en) * | 2015-11-27 | 2016-03-23 | 优渊商贸(上海)有限公司 | Image color recognition method and device |
Non-Patent Citations (2)
Title |
---|
林建忠: "《金融信息分析》", 30 April 2015, 上海交通大学出版社 * |
袁哲 等: ""缺陷石板的计算机优化排样研究"", 《石材》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111134613A (en) * | 2019-11-21 | 2020-05-12 | 明灏科技(北京)有限公司 | Image recognition-based orthokeratology lens fitting method and system |
CN112487950A (en) * | 2020-11-27 | 2021-03-12 | 上海建工四建集团有限公司 | Intelligent recognition system and method for defects of bare bricks |
CN113135368A (en) * | 2021-04-29 | 2021-07-20 | 怀化学院 | Intelligent garbage front-end classification system and method |
CN113135368B (en) * | 2021-04-29 | 2022-07-22 | 怀化学院 | Intelligent garbage front-end classification system and method |
CN115937085A (en) * | 2022-06-28 | 2023-04-07 | 哈尔滨学院 | Nuclear cataract image processing method based on neural network learning |
CN115434054A (en) * | 2022-10-28 | 2022-12-06 | 富尔美技术纺织(苏州)有限公司 | Multicolor fiber uniform mixing processing technology |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109800772A (en) | A kind of data identification method | |
CN111951237B (en) | Visual appearance detection method | |
CN108052880B (en) | Virtual and real lane line detection method for traffic monitoring scene | |
CN105718945B (en) | Apple picking robot night image recognition method based on watershed and neural network | |
Naji et al. | Skin segmentation based on multi pixel color clustering models | |
CN105184787A (en) | Identification camera capable of automatically carrying out portrait cutout and method thereof | |
CN109726717B (en) | Vehicle comprehensive information detection system | |
CN107301378B (en) | Pedestrian detection method and system based on multi-classifier integration in image | |
CN104408424B (en) | A kind of multi signal lamp recognition methods based on image procossing | |
CN108846359A (en) | It is a kind of to divide the gesture identification method blended with machine learning algorithm and its application based on skin-coloured regions | |
CN102819728A (en) | Traffic sign detection method based on classification template matching | |
CN103413147A (en) | Vehicle license plate recognizing method and system | |
CN110956099B (en) | Dynamic gesture instruction identification method | |
CN106909884B (en) | Hand region detection method and device based on layered structure and deformable part model | |
CN109949198B (en) | Wheat field boundary detection device and detection method | |
CN101364304B (en) | Shadow detection method based on color invariance and Gauss model | |
CN103440035A (en) | Gesture recognition system in three-dimensional space and recognition method thereof | |
CN106529461A (en) | Vehicle model identifying algorithm based on integral characteristic channel and SVM training device | |
CN103177445A (en) | Outdoor tomato identification method based on subsection threshold image segmentation and light spot identification | |
CN112560941A (en) | Coal and gangue identification method based on image detection | |
CN115035088A (en) | Helmet wearing detection method based on yolov5 and posture estimation | |
CN110110810B (en) | Squid quality grade identification and sorting method | |
CN107992799B (en) | Preprocess method towards Smoke Detection application | |
CN109829496B (en) | Physical measurement classification method and equipment | |
Jia et al. | Design of Traffic Sign Detection and Recognition Algorithm Based on Template Matching |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190524 |